Scholarlink

<p>Surrogate models enable rapid and computationally efficient slope stability evaluations for reliable engineering decision-making and risk assessment. This study addresses two key challenges in developing machine learning-based surrogate models for the rapid stability assessment of soil–rock mixture (SRM) slopes: (1) augmenting the dataset to increase training sample size and enhance data diversity, and (2) fusing heterogeneous SRM slope characteristics into structured input representations compatible with machine learning models. To address these challenges, an enhanced data augmentation method is employed, expanding the initial 400 generated datasets, thereby enriching the training set and improving the predictive generalization capability of the surrogate models. In parallel, three data fusion strategies are introduced to integrate heterogeneous geological and geotechnical data in SRM slope problems, transforming them into coherent and compatible input datasets for the surrogate model. Suitable surrogate training methods are then selected for each fusion strategy to generate three surrogate models. A comparative analysis is conducted to evaluate the performance improvements of the surrogate models before and after data augmentation across all three fusion strategies. The results confirm that data augmentation enhances prediction accuracy and generalization across the surrogate models. Among the three fusion strategies, the surrogate model employing the image–strength parameter fusion approach achieves an optimal trade-off between computational efficiency and predictive accuracy. This strategy is particularly well suited for rapid and reliable stability evaluations in practical on-site application.</p><p><b>Highlights</b><UnorderedList Mark="Bullet"> <ItemContent> <p>A reliable data fusion framework and scalable machine learning approach are proposed for surrogate modeling.</p> </ItemContent> <ItemContent> <p>A tailored data augmentation strategy enhances sample quality and diversity for failure prediction of soil–rock mixed slope.</p> </ItemContent> <ItemContent> <p>Construction efficiency and prediction accuracy of the surrogate model for SRM slope stability are jointly optimized.</p> </ItemContent> </UnorderedList></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Surrogate Model for Rapid Stability Prediction of Soil–Rock Mixture Slopes Based on Multi-source Data

  • 期刊论文

  • 发表日期：

    2026年3月2日

查看全文
  • Yuan Zhang,
  • Fusheng Zha,
  • Hui Yang,
  • Shan Wu,
  • Bo Kang,
  • Yixiao Pang,
  • Shixian Ren,
  • Junwei Su

摘要

Surrogate models enable rapid and computationally efficient slope stability evaluations for reliable engineering decision-making and risk assessment. This study addresses two key challenges in developing machine learning-based surrogate models for the rapid stability assessment of soil–rock mixture (SRM) slopes: (1) augmenting the dataset to increase training sample size and enhance data diversity, and (2) fusing heterogeneous SRM slope characteristics into structured input representations compatible with machine learning models. To address these challenges, an enhanced data augmentation method is employed, expanding the initial 400 generated datasets, thereby enriching the training set and improving the predictive generalization capability of the surrogate models. In parallel, three data fusion strategies are introduced to integrate heterogeneous geological and geotechnical data in SRM slope problems, transforming them into coherent and compatible input datasets for the surrogate model. Suitable surrogate training methods are then selected for each fusion strategy to generate three surrogate models. A comparative analysis is conducted to evaluate the performance improvements of the surrogate models before and after data augmentation across all three fusion strategies. The results confirm that data augmentation enhances prediction accuracy and generalization across the surrogate models. Among the three fusion strategies, the surrogate model employing the image–strength parameter fusion approach achieves an optimal trade-off between computational efficiency and predictive accuracy. This strategy is particularly well suited for rapid and reliable stability evaluations in practical on-site application.

Highlights

A reliable data fusion framework and scalable machine learning approach are proposed for surrogate modeling.

A tailored data augmentation strategy enhances sample quality and diversity for failure prediction of soil–rock mixed slope.

Construction efficiency and prediction accuracy of the surrogate model for SRM slope stability are jointly optimized.